Challenge
Automating our own recruitment, and making sure our recruiters actually use it.
We support our clients on the adoption of their data and AI products. We wanted to prove it first in-house, on a process that matters: our own recruitment.
The challenge was not technical. It came down to three requirements.
- Cover the whole process. From candidate sourcing to first outreach, without a break.
- Comply with GDPR. Processing candidates' personal data requires a clear compliance framework from the design stage.
- Achieve real usage. A tool that recruiters bypass is worthless, however capable it is.
The need: end-to-end automated recruitment that recruiters and business managers use every day.
Solution
Start from the recruiters' actual work, and measure adoption before building.
We treated the tool as a product, with its users and its indicators. Three principles structure the approach.
- The actual workflow first. The tool reproduces the way recruiters actually work, not a theoretical process.
- Adoption indicators set before development. Usage defines success, and we define that measurement from the start.
- Remove risks in order. The POC secures scoring and compliance, the MVP proves daily use, and industrialization comes next.
Set-up
A POC to secure, an MVP to prove, a foundation to last.
- Product scoping. Analysis of the recruiters' actual workflow and definition of adoption indicators before any line of code.
- POC. Validation of candidate scoring and of the GDPR compliance framework.
- MVP. Automated sourcing, scoring and outreach, placed in the hands of recruiters to prove daily use.
- Industrialization. Integration of the tool into Diametral's internal technical foundation.
A team of five: a product owner, a tech lead and three developers, supported by the CTO on the technical foundation. A ten-month project, from scoping in February 2026 to the rollout to all recruiters, planned for December 2026.
Tech stack: Python, JavaScript front end, containerized deployment; language model calls centralized in an internal LLM gateway, with the provider selected by configuration and each call logged with its prompt version, tokens and cost per feature; a profile data lake; integration with the Boond ATS.
Results
AI-assisted recruitment, where every decision rests on evidence.
A campaign starts with a single conversation. The assistant reads the job requirement sheet, suggests the missing fields while citing its source, and the recruiter accepts in one click. The filter is deterministic and explains each exclusion, criterion by criterion. Scoring qualifies high-potential profiles with the evidence found in the profile, and never excludes a candidate. The decision to contact and shortlist a candidate always remains with the recruiter.
We set the target before go-live: 20 qualified interviews per week per recruiter, compared with 6 to 10 before, by freeing up sourcing and administrative time. The rollout to all recruiters and business managers is planned for December 2026.
Above all, we apply to ourselves the method we offer our clients: adoption indicators defined before building, and risks removed step by step.
Expertises mobilisées
Savoir-faire
- Product management
- Adoption
- Scoring
- GDPR compliance
- Explainable AI
Technologies
- Python
- JavaScript
- LLM gateway
- Containers
- Boond ATS
Signature
Four mechanisms that make the difference.
- A method tested in-house. What we recommend to our clients, we first apply to our own business.
- Adoption as the measure of success. We set usage indicators before development, not after.
- Compliance from the POC. We validate the GDPR framework before scaling up.
- An AI that shows its evidence. Each suggested field cites its source, each exclusion is explained, and scoring never rules out a candidate.
For you
A data product adopted by those who use it.
This approach addresses four situations.
- You want to automate a business process end to end, not just an isolated task.
- Your internal tools exist but your teams do not use them.
- Your personal data requires a solid GDPR framework.
- You want to move from a POC to an industrialized product, with usage indicators.
We start from your teams' actual work and set the adoption indicators with you before building. The experts who design the product also put it into production and track its usage.
